docs/assets/foundation-pipelines/prompts.md

Foundation Pipeline Slide Source Notes

The current public assets are not generated concept art. They are restored high-resolution PNGs rebuilt from original direction-slide sources supplied by the project owner. The filename is kept as prompts.md because older public manifests and mirrors already link here as the provenance note.

Update on 2026-06-19: the latest supplied clean Spatial intelligence and Human-video world model PNGs are byte-identical to the committed source-slide cache and are published as 2560-pixel public images. A clean Vision-language-action PNG was then supplied and is now committed as the VLA source slide, replacing the temporary redraw fallback.

Track	Source	Enhanced public PNG
Spatial intelligence models	source-slides/spatial-intelligence-slide.png	spatial-intelligence-pipeline.png
Human-video world models	source-slides/human-video-world-model-slide.png	human-video-world-model-pipeline.png
Vision-language-action models	source-slides/vision-language-action-slide.png	vision-language-action-pipeline.png

Restoration is deterministic and local:

• Clean slide PNGs are used directly where available.
• EXIF orientation normalization.
• Autocontrast and moderate brightness/color/contrast correction.
• Lanczos resize to a 2560-pixel public width.
• Gentle sharpening and unsharp masking. The restoration script deliberately avoids hallucinated model results or non-source concept art. Technical task/training/evaluation scope is maintained in THREE_FOUNDATION_PIPELINES.md and docs/data/three_foundation_pipelines.json.